{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:OEVV2BURLATLHK6V5IN6IJ6LLR","short_pith_number":"pith:OEVV2BUR","schema_version":"1.0","canonical_sha256":"712b5d06915826b3abd5ea1be427cb5c4242327cc5f801e6a7da01603b623a2d","source":{"kind":"arxiv","id":"2607.08973","version":1},"attestation_state":"computed","paper":{"title":"SiFAR: Synchronization-Free All-Reduce for Low-Latency LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Abhishek Revinipati, Anish Saxena, Hritvik Taneja, Jae Hyung Ju, Moinuddin Qureshi, Neal C. Crago","submitted_at":"2026-07-09T22:34:00Z","abstract_excerpt":"The rise of reasoning models and agentic systems has made LLM token-generation latency a key bottleneck. Unlike chatbots, whose latency gains saturate at human reading speed, these systems generate intermediate reasoning tokens not consumed by humans. Thus, per-token latency directly determines end-to-end response time. Low-latency inference uses minimal batching, making token generation bandwidth-bound. Tensor Parallelism addresses this by sharding model weights across GPUs and loading them in parallel. However, scaling to more GPUs introduces All-Reduce overheads that grow with GPU count. Re"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2607.08973","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-07-09T22:34:00Z","cross_cats_sorted":[],"title_canon_sha256":"c4f44df06b469d96bf3b195577940d92e1f8c2c37e77cec9a9a0c156a925f847","abstract_canon_sha256":"a1a91aced1c5a39031c37e899a75ae26f5f21a35e6768f296523eb166a7a667a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-13T00:17:31.165850Z","signature_b64":"jlvxJ9Ik0XMIiwSO/5EFngRrIKFFdilXtOwLLEebkkkbjyLCm0VzcDr3BQoqWjDCGtInILcaNwfnlcgNA2gLCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"712b5d06915826b3abd5ea1be427cb5c4242327cc5f801e6a7da01603b623a2d","last_reissued_at":"2026-07-13T00:17:31.164024Z","signature_status":"signed_v1","first_computed_at":"2026-07-13T00:17:31.164024Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SiFAR: Synchronization-Free All-Reduce for Low-Latency LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Abhishek Revinipati, Anish Saxena, Hritvik Taneja, Jae Hyung Ju, Moinuddin Qureshi, Neal C. Crago","submitted_at":"2026-07-09T22:34:00Z","abstract_excerpt":"The rise of reasoning models and agentic systems has made LLM token-generation latency a key bottleneck. Unlike chatbots, whose latency gains saturate at human reading speed, these systems generate intermediate reasoning tokens not consumed by humans. Thus, per-token latency directly determines end-to-end response time. Low-latency inference uses minimal batching, making token generation bandwidth-bound. Tensor Parallelism addresses this by sharding model weights across GPUs and loading them in parallel. However, scaling to more GPUs introduces All-Reduce overheads that grow with GPU count. Re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.08973","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2607.08973/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2607.08973","created_at":"2026-07-13T00:17:31.165255+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.08973v1","created_at":"2026-07-13T00:17:31.165255+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.08973","created_at":"2026-07-13T00:17:31.165255+00:00"},{"alias_kind":"pith_short_12","alias_value":"OEVV2BURLATL","created_at":"2026-07-13T00:17:31.165255+00:00"},{"alias_kind":"pith_short_16","alias_value":"OEVV2BURLATLHK6V","created_at":"2026-07-13T00:17:31.165255+00:00"},{"alias_kind":"pith_short_8","alias_value":"OEVV2BUR","created_at":"2026-07-13T00:17:31.165255+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR","json":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR.json","graph_json":"https://pith.science/api/pith-number/OEVV2BURLATLHK6V5IN6IJ6LLR/graph.json","events_json":"https://pith.science/api/pith-number/OEVV2BURLATLHK6V5IN6IJ6LLR/events.json","paper":"https://pith.science/paper/OEVV2BUR"},"agent_actions":{"view_html":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR","download_json":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR.json","view_paper":"https://pith.science/paper/OEVV2BUR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.08973&json=true","fetch_graph":"https://pith.science/api/pith-number/OEVV2BURLATLHK6V5IN6IJ6LLR/graph.json","fetch_events":"https://pith.science/api/pith-number/OEVV2BURLATLHK6V5IN6IJ6LLR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR/action/storage_attestation","attest_author":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR/action/author_attestation","sign_citation":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR/action/citation_signature","submit_replication":"https://pith.science/pith/OEVV2BURLATLHK6V5IN6IJ6LLR/action/replication_record"}},"created_at":"2026-07-13T00:17:31.165255+00:00","updated_at":"2026-07-13T00:17:31.165255+00:00"}